Table 1.

Summary of the state-of-the-art of approaches discussed in this work.

Approach	Description	Applications	References
Spectrograms	The sound is recorded and then analyzed using spectrograms, searching for changes in the harmonic content of the signal.	Analysis of waggle dance, analysis of piping and tooting. Swarming detection. Measuring bees reaction to hornet attack.	[20,27,30,32,33]
Tone based sound synthesis	A loudspeaker or a shaker is placed inside the hive, different tones at different amplitudes and frequencies are generated and bees reaction is monitored.	Find the frequency at which bees react with movement cessation. Reproduce the harmonic generated by the queen bee, to stimulate a swarming.	[19,20,28]
Amplitude monitoring	Amplitude and envelope of the recorded signal is used to detect different behaviors.	Changes of the amplitude in different seasons and conditions. Measuring of SPL during waggle dance. Swarming detection.	[20,30,42]
Bees sound synthesis	The bees sound is firstly recorded and then analyzed and synthesized by means of a computer. The synthesized sound is then reproduced inside the colony and bees reaction is monitored.	Measure the response of worker bees to the synthesized queen bee sound.	[31]
Noise analysis	The recorded sound inside the colony is considered as a noise with a specific statistical behavior. Some statistical indicators are extracted from the sound, changes in the statistical indicators are related to specific colony behaviors.	Swarming detection and prediction.	[35]
Statistical indicator analysis	From the recorded sound, peak frequency, spectral centroid, bandwidth and root variance frequency are extracted. PCA is used to reduce the dimensionality of the indicators and finally SVM or LDA is used to classify the signals.	Detect the presence of Varroa destructor inside the colony.	[40]
Whooping detection	Precision accelerometer inside the colony are used to record the bees vibrations. Spectrograms of vibrations are cross correlated with a pulse signal to detect pulsed signals, LDA and PCA are then used to isolate whooping signals.	Measuring the variation of the whooping signal during different seasons and geographical locations.	[43]
Bees sound detection	The sound is acquired at the hive entrance. Spectrograms of the recorded sound are classified using different algorithms such as, CNNs, logistic regression, SVM, k-NN, one vs. rest and random forest.	Distinguishing the honey bee sound, from the background noise and the cricket chirping noise	[44]
LPC sound analysis	Sound acquired inside the hive is analyzed using LPC as features extraction algorithm. T-SNE algorithm is then used to reduce dimensionality, and finally SVM is used to classify the signals.	Queen bee presence detection.	[45]
HHT and MFCC analysis	Recorded sound inside the colony is analyzed using MFCCs and HHT as features. CNNs and SVM are then applied to classify the signals.	Queen bee presence detection, swarming detection.	[51,57,58]
MFCC analysis	MFCCs are estimated from the recorded signal. Lasso regularization is then used for dimensionality reduction and finally logistic regression algorithm is used for classification.	Queen bee presence detection.	[54]
Wavelet analisys	Wavelet transform is applied to the recorded signal to analyze the sound and detect different behavior.	Queen bee presence detection, swarming detection	[57,58]
MFCC and LPC analysis	MFCC and LPC are used as features, HMM and GMM are used as classifier.	Swarming detection.	[55]
Multimensional FFT	Two and three-dimensional spectrograms are generated starting from the sound recorded using accelerometers placed inside the colony. A discriminant function is then used to classify the signals and detect specific events using two different algorithms.	Swarming detection and swarming prediction.	[61]